Learning with Noisy Labels

By Nagarajan Natarajan, Inderjit S. Dhillon, Pradeep Ravikumar, Ambuj Tewari. 2013

1,313 citations recorded by the source corpus.

Subfields: Evaluation & Benchmarks, Statistical Learning Theory

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Evaluation & Benchmarks

In Statistical Learning Theory

Citations in the corpus

Prediction-Based Decisions and Fairness: A Catalogue of Choices, Assumptions, and Definitions (2018, background): “Measurement error in Y is sometimes called noisy labels [127], mislabeled data [17, 16], or unreliable oracles [154].”
On Robustness of Neural Architecture Search Under Label Noise (2020, background): “The third broad group of approaches aims to learn directly from noisy labels under the generic risk minimization framework and focus on noise-robust algorithms (Manwani and Sastry, 2013; Natarajan et al., 2013; Ghosh et al., 2017; Patrini et al., 2017; Zhang and Sabuncu, 2018).”
Detection and Continual Learning of Novel Face Presentation Attacks (2021, background): “This observation suggests a future direction for improving our algorithm is tackling the challenge of label pollution [18].”
Learning from Noisy Large-Scale Datasets with Minimal Supervision (2017, background): “ing [17], self-supervised learning [9, 24, 25, 31] and learning from noisy annotations [6, 23].”
Learning to Reweight Examples for Robust Deep Learning (2018, background): “Meanwhile, the noisy label problem has been thoroughly studied by the learning theory community (Natarajan et al., 2013; Angluin & Laird, 1988) and practical methods have also been proposed (Reed et al., 2014; Sukhbaatar & Fergus, 2014; Xiao et al., 2015; Azadi et al., 2016; Goldberger &…”
Co-teaching: Robust training of deep neural networks with extremely noisy labels (2018, methodology): “[30] proposed an unbiased estimator to provide the noise corrected loss approach.”
Generalized Cross Entropy Loss for Training Deep Neural Networks with Noisy Labels (2018, methodology): “[28] presented a way to modify any given surrogate loss function for binary classification to achieve noise-robustness.”
Learning from Noisy Labels with Distillation (2017, background): “This approaches extends the idea of unbiased loss function [14] in the traditional importance re-weighting framework.”
Using Trusted Data to Train Deep Networks on Labels Corrupted by Severe Noise (2018, background): “Accordingly, modeling, correcting, and learning with noisy labels has been well studied [14, 1, 2].”
Learning from Noisy Labels with Deep Neural Networks (2014, methodology): “An unbiased estimator for binary classiﬁcation is introduced in [12], where the cost function Although the extended model is trained on noisy labeled data, the original model should be used for testing.”
An Empirical Analysis of Backward Compatibility in Machine Learning Systems (2020, background): “Other studies have outlined the influence of class label noise [11, 19] and feature noise [14] on model behavior, informing a parallel line of work on increasing model robustness to noise [16, 25, 30, 37].”
On the Convergence of a Family of Robust Losses for Stochastic Gradient Descent (2016, methodology): “To generate the datasets with noisy labels, we follow the settings in [26,19].”
Detection of Adversarial Training Examples in Poisoning Attacks through Anomaly Detection (2018, methodology): “As suggested in [37] we have estimated these noise levels with a 5-fold cross validation procedure exploring the values given in the set {0, 0.”
On the Robustness of ChatGPT: An Adversarial and Out-of-distribution Perspective (2023, background): “Robustness threats exist in a wide range of scenarios: out-of-distribution (OOD) samples [55], adversarial inputs [15], long-tailed samples [60], noisy inputs [31], and many others.”
Normalized Loss Functions for Deep Learning with Noisy Labels (2020, background): “This approach improves robustness by modifying the loss function during training, based on label-dependent weights (Natarajan et al., 2013) or an estimated noise transition matrix that deﬁnes the probability of mislabeling one class with another (Han et al., 2018a).”
Classification without labels: learning from mixed samples in high energy physics (2017, background): “The problem of learning from unknown mixed samples can be shown to be mathematically equivalent to the problem of learning with asymmetric random label noise, where there have been recent advances [32, 40].”